| title | Stochastic Deep Supervision for Gallery-Consistent Early-Exit Face Recognition |
|---|---|
| authors | Guray Ozgur, Tahar Chettaoui, Eduarda Caldeira, Naser Damer, Fadi Boutros |
| date | 2026-11-02 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2611.09987 |
| affiliation | Fraunhofer IGD |
| upvotes | 2 (Hugging Face Daily Papers, 2026-11-12) |
| links | abs pdf hf code |
| cite | |
| related | Learning Steadily: Accumulating Relative Point Margin Scores for Face Image Quality Assessment Hard Sample-Conditioned Decorrelated Attention for Robust Face Recognition Tomatoes, Potatoes, and Onions: Questioning the Need for Faces in Face Presentation Attack Detection When Morphs Look Better: Face Image Quality as a Training-Free Signal for Morphing Attack Detection EXPL-FR: Explaining Face Recognition Models via Vision-Language Alignment |
| abstract | Vision Transformers (ViTs) perform strongly in face recognition (FR), yet their fixed-depth inference limits deployment in resource-constrained settings. We propose SD-ViT, a training framework that gives a usable exit at every block depth by making every depth produce discriminative embeddings, for budget-selected fixed-depth exits. Standard ViTs supervise intermediate layers only through backpropagation, so their intermediate representations are not directly optimized to form usable identity embeddings. SD-ViT adds a stochastic auxiliary signal: at each iteration a randomly selected intermediate block is passed through the same embedding head as the final layer and trained with a weighted margin-penalty softmax loss. The objective is thus stochastically sampled deep supervision through a shared classifier. Because both loss terms share that head and classifier, intermediate tokens are trained to map into the same embedding space as final-layer tokens. This matters for FR, where matching is by similarity: a probe processed at an early exit can be compared directly against full-depth gallery embeddings, with no gallery recomputation or adaptation layers. On established benchmarks, SD-ViT stays within 0.4 IJB-C points of the same backbone trained conventionally at full depth, while its early-exit probes recover over 96% of their own full-depth performance at 50% of the computational cost without modifying the reference gallery. A single verification threshold calibrated at full depth remains usable from quarter depth onward. |
bibtex
@misc{ozgur2026stochastic,
title={Stochastic Deep Supervision for Gallery-Consistent Early-Exit Face Recognition},
author={Guray Ozgur and Tahar Chettaoui and Eduarda Caldeira and Naser Damer and Fadi Boutros},
year={2026},
eprint={2611.09987},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2611.09987}
}